a = rand(1, 11);
N = length(a);
b = zeros(1,N - 1);
for i = 1:N-1
b(i) = a(i) + a (i + 1);
end
% Vectorized 1:
b2 = a(1:end-1) + a(2:end);
% Vectorized 2:
b3 = conv(a, [1,1], 'valid')
% [EDITED] % Vectorized 3: Working only for even length of a!
% [EDITED] b4 = sum(reshape(a, 2, [])); % Not the same result!
% Thanks John D'Errico.
isequal(b, b2, b3)
For a = rand(1,1e6) and 100 repetitions, Matlab 2018b needs:
% Elapsed time is 0.683647 seconds. loop
% Elapsed time is 0.874940 seconds. 1:end-1 + 2:end
% Elapsed time is 0.477227 seconds. conv
You see, that vectorization is not a standard trick for accelerating code. The memory access can take more time than the loop.
